SAN FRANCISCO – Artificial intelligence promises to make spacecraft increasingly resilient and capable of collecting data without waiting for instructions from ground controllers.
“Until now we've been limited in the way we've worked,” said Evana Gizzi, AI research leader at NASA's Goddard Space Flight Center. Space News“And there are so many things we want to do.”
Distributed missions, for example, where spacecraft work with landers and rovers to achieve common goals, will require autonomous capabilities. AI also paves the way for extensible mission architectures, allowing new spacecraft and sensors to join swarms in orbit.
“At NASA and in the aerospace industry in general, mission concepts are becoming more complex, meaning many of them can’t be done without AI,” said Gizzi, who earned a PhD in artificial intelligence from Tufts University.
Measuring methane
Still, introducing AI into NASA missions isn’t easy. Space mission planners tend to be risk-averse and understandably wary of untested algorithms.
To lower the barriers to introducing AI into spacecraft, NASA’s Goddard Space Autonomy and Resilience Laboratory (SPAR) created the onboard AI research platform, called OnAIR. OnAIR, an open-source software pipeline and cognitive architecture tool, is publicly available on the software development platform GitHub.
A prototype version of OnAIR was tested on NASA’s NAMASTE mission, which used a fleet of autonomous drones to measure methane distribution at Alaskan permafrost sites. (NAMASTE stands for Network for the Assessment of Methane Activity in Space and Terrestrial Environments.)
“OnAIR helped drones maximize data acquisition in areas of high scientific interest by providing a standard for data ingestion and processing involved in the NAMASTE software architecture, including autonomous measurements made by the drone-integrated Multifunctional Nanosensor Platform instrument,” Mahmooda Sultana, an instrument scientist at NASA’s Goddard Planetary Environments Laboratory with a PhD in chemical engineering from the Massachusetts Institute of Technology, said by email.
Tests on the ISS
OnAIR was also tested through the SpaceCube Edge-Node Intelligent Collaboration, or SCENIC. SpaceCube is a family of reconfigurable processors made from commercial, off-the-shelf, radiation-hardened components. In 2023, a SpaceCube was placed on the exterior of the International Space Station.
After completing its primary mission, including demonstrating the performance of commercial field-programmable gate arrays in space, the SPAR Laboratory overcame multiple challenges to perform an OnAIR demonstration.
“We originally thought we had a full year to prepare and upload OnAIR to SCENIC, but after two months we learned that SCENIC would be shut down earlier than expected,” James Marshall, a software engineer in NASA Goddard’s Science Data Processing Branch with a Ph.D. in computer science from George Washington University, said via email. As a result, Marshall and his colleagues were able to complete a year-long project in six months.
They also learned to work with SCENIC’s slow main processor. “The clock speed was 100 (megahertz) and the architecture is less common, so it was difficult to port the Python libraries we needed and performance was slow,” Marshall said.
Linking OnAIR to SCENIC’s existing core flight system posed another hurdle, which researchers overcame with knowledge from previous projects.
“The entire team (all three) had written software for SCENIC, so we were able to test everything on the SCENIC FlatSat and integrate the code easily,” Marshall said.
